White Paper: Inverse Multiplexing over ATM (IMA)

Asynchronous Transfer Mode (ATM) offers business benefits as a WAN technology and is on a steep growth curve, both in public...

Asynchronous Transfer Mode (ATM) offers business benefits as a WAN technology and is on a steep growth curve, both in public carrier networks and in private organisations

ATM across the WAN: changing the nature of networking

What are the challenges today for enterprise network managers seeking to maximise the effectiveness of their Wide Area Networks? Among the steepest: conquering complexity and reducing cost. With a growing number of remote offices to be linked to central sites, and with the explosive growth of online services and corporate intranets, extranets, and Internet access, businesses are depending on increased bandwidth for WAN access more than ever to carry out their daily operations. In addition, the increasing data bandwidth requirements and unforgiving delay constraints of real-time, interactive applications such as video streaming, group videoconferencing and telephony make it imperative that WAN links be as resilient and easy to manage as possible.

Over the past several years, Asynchronous Transfer Mode (ATM) has emerged as a technology of choice for reducing the complexity of WAN communications. A proven workhorse in LAN backbones, ATM offers many important benefits to the WAN. Among them are speed, scalability, traffic management and the ability to integrate LAN and WAN functions, binding voice, video, and data onto a single uniform protocol and design.

Inverse Multiplexing over ATM (IMA), a User-to-Network Interface (UNI) standard approved by the ATM Forum in 1997, raises ATM to an even higher level of WAN functionality and flexibility. In a nutshell, IMA specifies a transmission method in which ATM cells are fanned across several T1/E1 lines, then reassembled at the receiving end without loss of original ATM cell order. By enabling consolidated transport of the ATM protocol over cost-effective T1 and E1 lines, IMA extends ATM to all portions of the WAN, not just to locations where traffic is very high. Effectively, IMA delivers ATM to the masses.

ATM benefits in the WAN

You could think of IMA as the previously unknown factor that adds cost effectiveness into the ATM WAN equation. And because the result equals ATM benefits for all WAN users, not just those with very high traffic loads, it's worthwhile to quickly review ATM's WAN benefits.

Highly scalable bandwidth. ATM's biggest claim to fame is its speed: from 1.544Mbit/s to gigabit ranges, with 1.2Gbit/s (SONET OC-12) as the maximum customer premise bandwidth available. The benefit: incremental costs for incremental bandwidth, resulting in increased efficiency on high-traffic WAN links and an opportunity to "right-size" bandwidth needs even to very high user demand.

Network simplification through consolidation. ATM is the answer for combining applications that traditionally required different networks because of the different transport requirements of their traffic. This in turn lets network planners stop the proliferation of complex parallel networks: for example, one carrying data, another carrying voice, and another carrying video. ATM's ability to consolidate all types of traffic onto a single WAN link greatly reduces complexity and simplifies network management by eliminating these separately managed lines.

Bandwidth efficiency. Consolidation of diverse traffic types also lets network managers with high volumes of traffic fully utilise high-speed WAN links, instead of partially filling separate links with different types of traffic.

Quality of service. ATM offers bandwidth allocation based on user-defined needs and prioritisation, as well as load sharing of multiple technology types for guaranteed quality of service (QoS). ATM's traffic management controls enable seamless integration of voice, video and data while providing the separate management techniques required by each type of traffic.

Open connectivity. Because ATM is not based on a specific type of physical transport, it is compatible with all currently deployed physical networks. It can be transported over twisted pair, coax and fibre optics. And since ATM is a standard rather than a proprietary protocol, it can run on any vendor's standards-compliant products or be purchased from any carrier.

Excellent fault tolerance. ATM networks can be built with very high levels of fault tolerance at relatively low cost. IMA, for example, allows for load sharing and maximum network uptime.

ATM infrastructure availability. Service providers have invested heavily in the ATM infrastructure for reasons similar to those of enterprises: consolidation of traffic/back-bones, better bandwidth utilisation and so on. ATM can also be deployed as a private network built from leased lines such as T1/E1, T3/E3, or OC-3/STM-1. Taken in sum, ATM's capabilities-scalable bandwidth, network simplification, bandwidth efficiency, guaranteed QoS, open connectivity, fault tolerance, and infrastructure availability-make it invaluable for corporate WANs. ATM is also a stable WAN technology with an extensive public infrastructure. Up until now, the primary barrier to securing ATM benefits in the WAN has been the limited availability of carrier service.

Identifying the ATM WAN barrier

Despite all the benefits provided by ATM, the biggest deterrent to its deployment across the WAN has been a narrow choice of ATM WAN transmission speeds and the gap between low and mid-range services. This results in a cost structure that has made it difficult for the majority of businesses to "right-size" the number of WAN circuits to their network needs. Let's look at the options that were available to network planners prior to the development of IMA technology.

Option 1: T1/E1

At 1.544/2.048Mbit/s, T1/E1 lines are cost effective and widely accessible. But at the edge of the network, moving out to the WAN, a significant portion of businesses need to scale beyond a single T1/E1 link-especially with higher-speed LAN technologies, soon to include Gigabit Ethernet, which will put increasing bandwidth pressure on WAN links.

Option 2: T3/E3

At 44.736/34.368Mbit/s, or about 17 to 28 times the capacity of T1/E1, T3/E3 services are the next choice up. T3/E3 utilises the high bandwidth capacity of ATM, but at a huge increase in price, with particularly high costs in Europe and Asia. In fact, in most cases the charges for T3/E3 services in these markets are prohibitive even to the main corporate office. What's more, T3/E3 services are not widely available. The costs of OC-3 or STM-1 155Mbit/s fibre services, where available, are also beyond the range of most corporate budgets.

Option 3: Incremental T1/E1

Given the bandwidth disparity between the first two options, network planners who lack enough traffic to justify T3/E3 but whose networks have outgrown a single T1/E1 circuit must allocate WAN bandwidth in T1/E1 increments. Depending on location and carrier, multiples of T1/E1 provide a flexible, cost-effective solution for networks that require up to four or up to eight T1/E1 circuits. If an organisation needs to add capacity above four or eight circuits (however, it is more cost-effective to use T3/E3 rather than purchase additional T1/E1 trunks for the majority of businesses) traffic WAN bandwidth requirements lie above a single T1/E1, but below the price point crossover to T3/E3.

While this approach delivers flexible bandwidth, in terms of delivering ATM benefits to the WAN, there are severe drawbacks to using incremental T1/E1 circuits ( even if the price is right. Because of the relatively small capacity of each circuit, for organisations whose traffic needs surpass a single T1/E1 line, incremental T1/E1 forfeits a defining benefit of ATM: the ability to aggregate and manage traffic across circuits. And without this traffic consolidation, circuits multiply and the benefits of ATM bandwidth efficiency and network simplification are lost.

Enter IMA: affordable, accessible ATM for the WAN

With the introduction of IMA, however, the option of incremental T1/E1 looks a great deal more attractive to organisations that have outgrown a single T1/E1 but are beneath the price point crossover to T3/E3. In fact, introducing IMA to the network entirely eliminates the drawbacks of incremental T1/E1, restoring the core ATM benefits of traffic consolidation, bandwidth efficiency, and network simplification. In addition, IMA introduces the additional benefit of improved fault tolerance through traffic management and link control.

What exactly does IMA do? As an example, IMA can take traffic from a relatively high-bandwidth connection, such as a campus ATM backbone running at 155 Mb/s, and spread it across multiple T1/E1 WAN circuits.

The aggregate bandwidth of any number of these T1/E1 lines (nxT1/E1) determines the rate of the ATM connection. IMA lets network managers utilise voice, video and data WAN bandwidth, while offering all the benefits of ATM at a more affordable T1/E1 cost. What's more, because IMA lets network planners provision bandwidth in T1/E1 increments, it is possible to increase or decrease bandwidth based on users' needs. When multiple T1/E1 circuits are multiplexed, they appear to customer equipment as one logical pipe.

( Transport of a single ATM cell stream at rates between T1/E1 and T3/E3, taking advantage of cost-effective bandwidth at sub-T3/E3 rates

( Provisioning of bandwidth in T1/E1 increments, which lets network planners increase or decrease bandwidth based on need

( Bandwidth consolidation across T1/E1 link groups, leading to more efficient use of circuits

( Automatic and transparent adjustment to accommodate added/restored and deleted/failed T1/E1 links, minimising provisioning and maintenance

( Transparent transport of the ATM layer and higher layers, which preserves cell order and ATM traffic management techniques and makes IMA compatible with the existing ATM architecture

How does IMA work?

To understand the benefits provided by IMA, it's important to understand what the standard provides and how inverse multiplexing functions. Essentially, IMA works by distributing the cells in ATM cell streams over multiple T1/E1 physical links. Each link is a standard T1/E1 ATM UNI, and cells are placed on the links on a per-cell basis, using a cyclic round-robin approach. For example, the first cell is sent on the first T1/E1 circuit, the second on the second circuit, and so forth. Control information is also sent so that the status of each link and the quality of the connection can be determined and automatically corrected. Cells are then recombined by the IMA device at the receiving end of the stream.

Since the IMA access device at the receiving end requires a steady stream of cells to correctly recreate the original stream, the sending device introduces filler cells to keep the round-robin process at both ends in sync whenever there is a lull in traffic. To reduce bandwidth consumption, IMA removes idle and unassigned cells from the original stream and reinserts them at the receiving end. IMA's inverse multiplexing is transparent to the application and to the rest of the network, because cell order and format is retained and the T1/E1 delay variations within each IMA group are compensated for by software buffers in IMA equipment.

The IMA UNI is carried on top of a T1 or E1 ATM physical interface, performing inverse multiplexing using the IMA Control Protocol (ICP). The ATM Forum defines how ATM cells are mapped onto physical layer media. In the case of IMA, however, the cell-based control protocol aggregates the WAN links.

In addition to inverse multiplexing, the IMA specification includes sections on link management, connection to cell sources, cell function and cell adaptation to non-ATM data, and unit (network device) management. These components define IMA at a number of ATM protocol layers. The specification also includes a discussion of cell synchronisation for IMA operation in networks containing multiple clocks.

Getting the most out of IMA

What are the capabilities you should expect when evaluating IMA solutions? First and foremost, look for devices that don't just implement IMA, but are engineered around the IMA standard. IMA products should be able to take advantage of ATM's sophisticated capabilities for traffic management, fault tolerance and legacy equipment interoperability through ATM Forum standards. They should also be able to manage bandwidth to provide guaranteed delivery of voice and video, while buffering data, in a single integrated network. And finally, vendor solutions should be offered in a range of configurations to ensure proper scalability. This allows efficient utilisation of WAN links, greater network flexibility and optimal per-site return on investment.

Let's take a closer look at what's required to support ATM traffic management and see why this requirement is so important. To be successful in the WAN, IMA devices must be able to use carrier services, plus combine diverse carriers' services in a single network to guarantee route diversity. To do this, the IMA vendor's products must support ATM traffic shaping and class of service to match the carrier service contract. Traffic shaping is the ability to meter bandwidth onto each WAN circuit so that it never exceeds the bandwidth purchased from the service provider.

In addition, since IMA devices must operate across multiple networks, they must also operate under multiple network clocks, where non-synchronous circuits are routed through different paths and more than one timing domain. Each IMA receiving device must be able to implement controlled frame slippage to compensate for the timing differences between circuits and master clocks, and work in a hybrid (mixed public/private) network.

The vendor's products should also support the capabilities of IMA fault tolerance and provide robust WAN availability features. The product should be able to monitor IMA T1/E1 links for performance and take automatic action to "heal" the network when a link is broken. For advanced fault tolerance, each T1/E1 link should deliver data into its own buffer, which needs to be sufficient to tolerate its potential component of T1/E1 link differential delay variation (the greatest difference in delay between any two links in the IMA bundle). If T1/E1 errors or excessive IMA link delay variation occurs, the IMA device should be able to identify and automatically remove the bad link from the group. The rest of the links should survive this failure and continue to pass network traffic. Insist on IMA solutions that demonstrate this level of fault tolerance.

IMA products must also interoperate with legacy technologies. ATM access products with IMA should be able to interface to data, voice, and video networks without expensive upgrades to existing equipment, and without compromising the services those networks provide to end users. Within the IMA specification, legacy interoperability is provided via internetworking standards such as those called out by the various standards bodies (ATM Forum, ITU-T, Frame Relay, etc.). Legacy sources such as Ethernet, Frame Relay, serial applications, circuit switches and others should be supported. For example, an Ethernet or Frame Relay interconnection lets an existing router or routed application with no ATM interface connect to an ATM access device, where its data traffic may be combined with voice and video and then sent on to the WAN using IMA.

And finally, an important caution: as you evaluate IMA products, be wary of vendors who offer a narrow range of products or a single product that promises to accomplish everything. Instead, select a product line-up that is designed for flexibility of solutions, one that offers a wide set of product options and an opportunity for cost savings. Many vendors with a single solution oblige their customers to overpurchase, which actually forfeits the scalability and flexibility at the core of the IMA standard. Big switches for small jobs, for example, are cost-excessive up front as well as over the long term due to the price of maintenance, upgrades, and administration, and are likely to waste bandwidth by using expensive, high-capacity switch slots for low line rates such as T1/E1.

Compiled by Mike Burkitt

Randy Brumfield ( 1997 3Com Corporation

Read more on Voice networking and VoIP